# HG changeset patch
# User Michael Spacefalcon <msokolov@ivan.Harhan.ORG>
# Date 1384230527 0
# Node ID 277fd7b971f0589adae5fbc00bec99ce5fbb6557
# Parent  fdfb57a1c5fe55c564704a302a5701d021177192
some success in finding familiar TI code in moko11 and Pirelli fw binary images

diff -r fdfb57a1c5fe -r 277fd7b971f0 moko11
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/moko11	Tue Nov 12 04:28:47 2013 +0000
@@ -0,0 +1,28 @@
+The Init_Target() function in the TCS211 code from Sotovik (which sits in a
+binary lib with no source!) programs nCS0 and nCS1 memory timings with WS=3.
+We would like to determine whether or not the moko11 firmware does the same
+thing.  We have no linker map file for moko11, so we have to dig around in the
+binary and try to match the code against known objects.
+
+In the Sotomodem version of Init_Target(), at offset 0x60 from the beginning of
+the function there is a BL instruction calling $CLKM_InitARMClock, and this call
+is immediately followed by the code that sets up the memory timings.
+
+Let's see what we can find in the moko11 binary image:
+
+0012D4:	RESET vector jumps here
+010000: the code here appears to fully match the .inttext section of
+	TI's int.obj
+010058:	appears to be the _INT_Initialize entry point
+	(seems to be the same for all TI firmwares of that era)
+010268:	b 0x1e8364, should be a jump to the _INC_Initialize veneer
+1D1E48:	first function called from Application_Initialize, should be
+	Init_Target()
+	Matches the Sotomodem version of Init_Target() indeed,
+	including the memory timing setup!
+1E72B0:	Expected start of $INC_Initialize, appears to match
+1E72F4:	bl 0x1e81fc, should be calling Application_Initialize()
+1E81FC:	Expected start of Application_Initialize(), contains 6 calls indeed
+1E8364:	looks like an ARM->Thumb call veneer indeed
+1E8370:	Thumb code begins, does bl 0x1e72b0
+1E8378:	back to ARM, veneer return
diff -r fdfb57a1c5fe -r 277fd7b971f0 pirelli/firmware
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/pirelli/firmware	Tue Nov 12 04:28:47 2013 +0000
@@ -0,0 +1,23 @@
+Following on the success of our match of moko11 disassembly against some known
+objects (see ../moko11), let's try doing the same thing with Pirelli's fw.
+
+Let's see if the code in Pirelli's fw at 0x40000 matches .inttext from TI's
+int.obj: so far, so good!  Let's see how far we can get:
+
+040000:	beginning of match with .inttext in TI's int.obj
+040268:	b 0x3f6b40, should be a jump to the _INC_Initialize veneer
+3BB7D4:	first function called from Application_Initialize()
+	the logic of Init_Target() is recognizable, but it's a modified
+	version, not the same object blob as we have
+	the setup of memory timings matches that done by OsmocomBB!
+3F11F8:	this should be Application_Initialize()
+	differences begin: instead of 6 function calls, there are 12,
+	with one of them conditionalized on the return value of the previous
+3F3E74:	expecting to see $INC_Initialize here - yes!
+3F6B40:	looks like an ARM->Thumb call veneer indeed
+3F6B4C:	Thumb code begins, does bl 0x3f3e74
+3F6B54:	back to ARM, veneer return
+
+data objects:
+
+01775048:	INC_Initialize state variable